Sementris

Semantris is a set of AI-powered word association games that utilize machine learning to predict related words based on user input. The game includes two modes: 'Verse by Verse,' where players co-create poetry with suggestions from famous poets, and 'Blocks' and 'Arcade,' which challenge players to come up with clues for word associations. These games serve as experiments in understanding and utilizing conversational AI technology.

Uploaded by

dlunker11

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

115 views4 pages

Sementris

Uploaded by

dlunker11

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 4

Semantris

Google’s AI generated game

Darsh Lunker
VIBGYOR RISE Mindspace
Semantris
Intro:

Semantris is a set of word association games powered by machine learning. By training on

billions of conversations from the internet, the AI has learned how to predict which words,
phrases, and even sentences might come next in a conversation.
Both games use the same AI, but provide different ways of interacting with it. When you
enter a clue, the AI looks at all the words in play and chooses the ones it thinks are most
related.
The games are experiments using technology that is still in development. Play around with
different types of clues and see what references the AI understands best.

How It Works
Ever wondered how a computer can understand everyday language when you talk to it?
The key lies in machine learning—a powerful technique that allows computers to learn
patterns from data. Our approach involved training an AI system on billions of lines of real
human dialogue to help it understand how conversations naturally unfold.
Once trained, the AI doesn’t rely on memorized responses. Instead, it learns to predict which
replies are most likely to come next in a conversation. In the demos you’re seeing, the AI
treats whatever you type as the beginning of a dialogue and searches through a wide pool of
potential replies to find the ones that most naturally follow.
The process is built around a concept known as machine learning. According to Google’s
Machine Learning Glossary:
"Machine learning is a program or system that builds (trains) a predictive model from input
data."
What does that mean in practice?
• Input data: We use billions of pairs of sentences where each pair consists of a
statement and its response.
• Prediction: The AI learns to anticipate what a likely response would be based on all
the examples it has seen.
• Model: After training, we get a system—a model—that can make smart predictions
by choosing the most appropriate response from a set of options.
Semantris
Verse by Verse
Verse by Verse is a unique poetry experience that lets you co-create with some of America's
most iconic poets—like Walt Whitman, Emily Dickinson, and Edgar Allan Poe. As you
compose, these literary giants offer suggestions for the next line of your poem, guiding and
inspiring your creative journey.
Verse by Verse
This tool brings two advanced AI models together:
1. Generative Model – Trained on classic American poetry, this model has learned how
to write new verses in the distinct style of each featured poet.
2. Semantic Model – This model understands the meaning of your lines and suggests
verses that naturally and meaningfully follow them. It uses the same underlying
language comprehension technology as the rest of our conversational applications.
Each poet has a unique voice, and you’ll likely find the experience most engaging when you
treat them as creative partners. You can choose to use their suggestions exactly as offered,
tweak them to better suit your vision, or simply let them inspire your own original lines.
Keep in mind, Verse by Verse is an experiment—an exploration of how machine learning can
serve as a creative collaborator. While it's designed to inspire and assist, it may occasionally
produce unexpected or quirky outputs. If you encounter issues, feel free to report them
using the feedback tool (look for the exclamation mark in a speech bubble at the top right
corner).

Blocks:
Semantris
"Blocks" explores what types of word associations the AI understands best.
When you submit a clue, the AI will look at all the blocks and remove the one it thinks is
most related. If it’s touching blocks that share a color, they’re removed as well. You can earn
more powerful blocks by destroying larger groups!
Take as long as you'd like to come up with creative clues and see which ones the AI
understands best. Try using slang, lists of examples, pop culture references, and even full
sentences.

Arcade:

"Arcade" challenges you to quickly come up with word association clues that the AI can
understand.
When you submit a clue, the AI looks at all the words in the list and sorts the ones it thinks
are most related to the bottom. Give great clues multiple times in a row to earn a streak
bonus that clears the whole list!
Type fast and use short phrases that accurately point to the targeted words. If you’re stuck
on a tricky word, try a different approach by using examples, categories, synonyms,
antonyms, or slang.
Of course! Here's the text from the image: